Hacker News Reader

Comment by reqo | Hacker News Reader

Another probable bad side effect of AI will be that a generation of developers will become dependent on copilot/chatgpt to be able to write code! They will definitely perform worse than before llm era without the help of a llm!

View on HN

Same could be said about developing with Notepad as opposed to modern IDEs though.

Did modern IDEs start writing code for you when you made the jump from Notepad?

In a lot of ways, yeah. Autocomplete is an absolute game changer and one of the biggest reasons that static typing came back in vogue and crushed dynamic languages. Massive productivity gains when you can ctrl+space and see all of the methods available on a variable, or use a hotkey to figure out if the method you want to call is `do_foo_bar` or `do_foo_and_bar`.

I'm not sure I buy that argument about dynamic languages. You are probably right in that auto complete played a role, but for me working on a pure node js app back in 2014ish, it was a nightmare of runtime issues that type checking would have eliminated. The productivity gain in a typed language was not having to write unit tests for things that static typing does for free.

I agree with you, autocomplete definitely isn't the only reason for static typing making a comeback and dynamic language popularity dying off (and the big legacy names like Javascript and Python have Typescript and mypy/pyright to transform them into a facsimile of one). I think a lot of it has to do with maturity, where a lot of "move fast and break things" devs who loved untyped Node/PHP/Python got burned by bugs and impossible to comprehend legacy codebases and slowly realized the value of static typing. But that type of understanding takes years. A novice programmer can see the value of autocomplete and not having to check the docs or stackoverflow for the name of a method pretty much instantly.

autocomplete accurately took care of a lot of the boilerplate; that low-hanging fruit has been picked. AI is not doing the same IME, the pay-off has been a lot slower coming, and sometimes it's making the work more painful.

I do agree. It's hard to get an unbiased assessment of how many developers are actually using generative AI (copilot/chatgpt) for work, and in what capacity they do use them. Anecdotally, at my org of ~40 devs, we encouraged everyone to try out Copilot and let us know if they wanted a full license; only 2 people took up the offer, and they use it either for generating unit tests or translating English data logic to pandas syntax (which it does seem quite good at!)

This argument rears its head every time the state of the art in developer tools makes a credible threat to improve.

And yet relatively few developers use C, assembly, or machine language as their daily drivers. I've even met accomplished developers who don't know how to write a compiler!

It's fine for developers to become dependent on tooling that makes them more efficient. This is a good thing, and we want more of this!

I think it's a bit different discussion. I'm 100% agree to use any tools including LLMs, but LLMs may cause not good affect for ecosystem.

New programming language, syntax, and API don't get LLMs support until they learned it (or given as a context). And perhaps new things less being used because of LLMs don't use it. Less usage in the wild means LLMs learn less from the nature. I wish it doesn't stop evolving programming (and other things).

You're going further to "whether LLMs can be useful when programming in a new programming language." Which I will leave to developers using those new languages.

I'm saying more that every time developers develop/find a tool that makes them more productive, if it doesn't look like the Old Ways, there will be other developers to crap on them and say they don't know enough about the Old Ways. Which may or not be true but isn't always relevant given the size of the space that comprises modern programming.

This is why there are still people who deride programmers who use IDEs instead of using tools that were invented nearly 50(!) years ago. I'm not saying either way is a fit for every use, but the attitude that developers should eschew attempts to improve the state of our tools is toxic.

That's probably what everyone said when auto-complete was invented and IDEs were used. The actual code written in most cases isn't so important, the architecture, experience in building systems and debugging skills across a whole stack are what counts.

But when would they not have the help of an LLM? We are all dependent on the tools we use. That's not a bad thing.

The biggest risk is that it perhaps becomes more difficult to sift out bad developers from good ones, because LLMs let bad developers "cheat" more easily. But that's not a new problem, just a new era of the same problem.

If the best coding LLMs end up monopolized as much as most areas of the tech industry it will be very easy to lose access. Take mobile app development for example. If you piss off the wrong person or algorithm at Apple or Google then poof, half the market will evaporate for you.

Corporate LLMs have and will monitor every single prompt entered and if the company doesn't like you for any reason, related or unrelated to your coding they will ban your access with no recourse.

If open source LLMs are able to compete with the state of the art long term then I'm a lot more optimistic, but I'm skeptical they will be able to keep pace given the immense technical, financial, and lobbying resources the big players are certain to invest in the coming years.

I said the same thing when they made compilers and cobol

All skills that the AI facilitates will be like this until all we can do is operate the AI

Exactly. We don't need an abacus. We need a human able to input the right variables into a computer and get the result.

Reply on news.ycombinator.com